Re: [nsp] about routing

From: Dorian R. Kim (dorian@blackrose.org)
Date: Wed Feb 04 1998 - 16:48:35 EST


On Wed, Feb 04, 1998 at 10:29:12PM +0100, Havard.Eidnes@runit.sintef.no wrote:
> Well, from what I hear there are good reasons why cisco is
> implementing use of a distributed forwarding table for use in their
> high-end gear (known as "Cisco Express Forwarding"/CEF or "FIB
> switching"). I guess some of the reasons are that cache invalidations
> cost too much because the main CPU can be a bottleneck in forwarding

Yes. Cache maintenance, in theory and in practice, is a very expensive
process. The cost of cache miss compounds that even more. Building a
full forwarding table and distributing that table to multiple processors
is a much more scalable design.

> decisions. I hear that in the Cisco 12000 boxes the main CPU can't do
> any packet forwarding at all; all the forwarding decisions have to be
> taken by the "line cards".

This was intentional. It's better to remove the temptation before some
decides to use the RP for forwarding.

-dorian



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:15 EDT